Electricity Consumption and Carbon Factor Forecasting for High-Energy-Consuming Enterprises Using Temporal Convolutional Network

Accurate forecasting of electricity consumption and carbon emission factors (CEF) is critical for high-energy-consuming enterprises aiming to achieve energy efficiency and carbon reduction goals. This study proposes a forecasting approach based on Temporal Convolutional Networks (TCN) to predict electricity usage and electricity carbon factors. The TCN model’s performance is compared against three widely used deep learning architectures: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Transformer. Real-world electricity usage data from energy-intensive enterprises is utilized to train and evaluate each model. Experimental results show that the TCN-based model outperforms LSTM, GRU, and Transformer in both prediction accuracy and training stability. Specifically, the TCN approach reduces the Mean Absolute Percentage Error (MAPE) by 12.7% on electricity consumption forecasting and by 15.3% on carbon factor prediction, compared to the best-performing alternative model. These findings highlight the effectiveness and robustness of TCNs as a superior solution for time series forecasting in the context of industrial energy management.